0-gather.org

gather

People use gather to mean bringing things or people together in one place. A family gathers for dinner. Someone gathers sticks for a fire. It is a plain, social word about collecting or assembling.

What a model may hear

collect data data engineering, analytics
may assume the user wants to scrape, query, or aggregate datasets from multiple sources
gather operation parallel computing, MPI
may assume a distributed process where one node collects results from all others, or write code using MPI_Gather
gather instruction GPU shaders, vector processing
may assume the user wants a SIMD gather operation that loads non-contiguous memory elements into a register
gather step machine learning pipelines
may assume collecting metrics, gradients, or model outputs from distributed training workers

Where people and models part ways

“gather all the customer feedback”

Meant: collect emails and survey responses in one document

May be taken as: build a web scraper or set up a real-time data pipeline

Say instead: “collect all customer feedback into one list from files I already have”

“gather the results from each team”

Meant: ask each team to send me their report

May be taken as: implement an MPI gather or distributed computing routine

Say instead: “ask each team for their report and compile them into one summary”

“gather the scattered data points”

Meant: organize numbers that are spread across a spreadsheet

May be taken as: use a GPU gather instruction or write vectorized assembly

Say instead: “organize numbers from different columns into one sorted column”

“let's gather online”

Meant: have a video call

May be taken as: assume the user wants to build or use a platform named Gather or Gather.town

Say instead: “meet on a video call”

Tips

Often confused with

collect
more neutral, less likely to trigger distributed computing
aggregate
strongly signals data summarization, not social assembly
assemble
closer to physical or social gathering, but can mean code compilation
join
specific to database tables or meetings, rarely parallel computing
reduce
a paired operation in parallel computing, not social
scrape
gathering from the web specifically, often unwanted